Pileus Git - ~andy/linux/blob - net/ipv4/inet_connection_sock.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Support for INET connection oriented protocols.
   7  *
   8  * Authors:     See the TCP sources
   9  *
  10  *              This program is free software; you can redistribute it and/or
  11  *              modify it under the terms of the GNU General Public License
  12  *              as published by the Free Software Foundation; either version
  13  *              2 of the License, or(at your option) any later version.
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/jhash.h>
  18
  19 #include <net/inet_connection_sock.h>
  20 #include <net/inet_hashtables.h>
  21 #include <net/inet_timewait_sock.h>
  22 #include <net/ip.h>
  23 #include <net/route.h>
  24 #include <net/tcp_states.h>
  25 #include <net/xfrm.h>
  26
  27 #ifdef INET_CSK_DEBUG
  28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  30 #endif
  31
  32 /*
  33  * This struct holds the first and last local port number.
  34  */
  35 struct local_ports sysctl_local_ports __read_mostly = {
  36         .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock),
  37         .range = { 32768, 61000 },
  38 };
  39
  40 unsigned long *sysctl_local_reserved_ports;
  41 EXPORT_SYMBOL(sysctl_local_reserved_ports);
  42
  43 void inet_get_local_port_range(int *low, int *high)
  44 {
  45         unsigned int seq;
  46
  47         do {
  48                 seq = read_seqbegin(&sysctl_local_ports.lock);
  49
  50                 *low = sysctl_local_ports.range[0];
  51                 *high = sysctl_local_ports.range[1];
  52         } while (read_seqretry(&sysctl_local_ports.lock, seq));
  53 }
  54 EXPORT_SYMBOL(inet_get_local_port_range);
  55
  56 int inet_csk_bind_conflict(const struct sock *sk,
  57                            const struct inet_bind_bucket *tb, bool relax)
  58 {
  59         struct sock *sk2;
  60         struct hlist_node *node;
  61         int reuse = sk->sk_reuse;
  62         int reuseport = sk->sk_reuseport;
  63         kuid_t uid = sock_i_uid((struct sock *)sk);
  64
  65         /*
  66          * Unlike other sk lookup places we do not check
  67          * for sk_net here, since _all_ the socks listed
  68          * in tb->owners list belong to the same net - the
  69          * one this bucket belongs to.
  70          */
  71
  72         sk_for_each_bound(sk2, node, &tb->owners) {
  73                 if (sk != sk2 &&
  74                     !inet_v6_ipv6only(sk2) &&
  75                     (!sk->sk_bound_dev_if ||
  76                      !sk2->sk_bound_dev_if ||
  77                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  78                         if ((!reuse || !sk2->sk_reuse ||
  79                             sk2->sk_state == TCP_LISTEN) &&
  80                             (!reuseport || !sk2->sk_reuseport ||
  81                             (sk2->sk_state != TCP_TIME_WAIT &&
  82                              !uid_eq(uid, sock_i_uid(sk2))))) {
  83                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
  84                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
  85                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
  86                                         break;
  87                         }
  88                         if (!relax && reuse && sk2->sk_reuse &&
  89                             sk2->sk_state != TCP_LISTEN) {
  90                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
  91
  92                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
  93                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
  94                                         break;
  95                         }
  96                 }
  97         }
  98         return node != NULL;
  99 }
 100 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
 101
 102 /* Obtain a reference to a local port for the given sock,
 103  * if snum is zero it means select any available local port.
 104  */
 105 int inet_csk_get_port(struct sock *sk, unsigned short snum)
 106 {
 107         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 108         struct inet_bind_hashbucket *head;
 109         struct hlist_node *node;
 110         struct inet_bind_bucket *tb;
 111         int ret, attempts = 5;
 112         struct net *net = sock_net(sk);
 113         int smallest_size = -1, smallest_rover;
 114         kuid_t uid = sock_i_uid(sk);
 115
 116         local_bh_disable();
 117         if (!snum) {
 118                 int remaining, rover, low, high;
 119
 120 again:
 121                 inet_get_local_port_range(&low, &high);
 122                 remaining = (high - low) + 1;
 123                 smallest_rover = rover = net_random() % remaining + low;
 124
 125                 smallest_size = -1;
 126                 do {
 127                         if (inet_is_reserved_local_port(rover))
 128                                 goto next_nolock;
 129                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 130                                         hashinfo->bhash_size)];
 131                         spin_lock(&head->lock);
 132                         inet_bind_bucket_for_each(tb, node, &head->chain)
 133                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 134                                         if (((tb->fastreuse > 0 &&
 135                                               sk->sk_reuse &&
 136                                               sk->sk_state != TCP_LISTEN) ||
 137                                              (tb->fastreuseport > 0 &&
 138                                               sk->sk_reuseport &&
 139                                               uid_eq(tb->fastuid, uid))) &&
 140                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 141                                                 smallest_size = tb->num_owners;
 142                                                 smallest_rover = rover;
 143                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
 144                                                     !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 145                                                         snum = smallest_rover;
 146                                                         goto tb_found;
 147                                                 }
 148                                         }
 149                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 150                                                 snum = rover;
 151                                                 goto tb_found;
 152                                         }
 153                                         goto next;
 154                                 }
 155                         break;
 156                 next:
 157                         spin_unlock(&head->lock);
 158                 next_nolock:
 159                         if (++rover > high)
 160                                 rover = low;
 161                 } while (--remaining > 0);
 162
 163                 /* Exhausted local port range during search?  It is not
 164                  * possible for us to be holding one of the bind hash
 165                  * locks if this test triggers, because if 'remaining'
 166                  * drops to zero, we broke out of the do/while loop at
 167                  * the top level, not from the 'break;' statement.
 168                  */
 169                 ret = 1;
 170                 if (remaining <= 0) {
 171                         if (smallest_size != -1) {
 172                                 snum = smallest_rover;
 173                                 goto have_snum;
 174                         }
 175                         goto fail;
 176                 }
 177                 /* OK, here is the one we will use.  HEAD is
 178                  * non-NULL and we hold it's mutex.
 179                  */
 180                 snum = rover;
 181         } else {
 182 have_snum:
 183                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 184                                 hashinfo->bhash_size)];
 185                 spin_lock(&head->lock);
 186                 inet_bind_bucket_for_each(tb, node, &head->chain)
 187                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 188                                 goto tb_found;
 189         }
 190         tb = NULL;
 191         goto tb_not_found;
 192 tb_found:
 193         if (!hlist_empty(&tb->owners)) {
 194                 if (sk->sk_reuse == SK_FORCE_REUSE)
 195                         goto success;
 196
 197                 if (((tb->fastreuse > 0 &&
 198                       sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 199                      (tb->fastreuseport > 0 &&
 200                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 201                     smallest_size == -1) {
 202                         goto success;
 203                 } else {
 204                         ret = 1;
 205                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 206                                 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
 207                                      (tb->fastreuseport > 0 &&
 208                                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 209                                     smallest_size != -1 && --attempts >= 0) {
 210                                         spin_unlock(&head->lock);
 211                                         goto again;
 212                                 }
 213
 214                                 goto fail_unlock;
 215                         }
 216                 }
 217         }
 218 tb_not_found:
 219         ret = 1;
 220         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 221                                         net, head, snum)) == NULL)
 222                 goto fail_unlock;
 223         if (hlist_empty(&tb->owners)) {
 224                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 225                         tb->fastreuse = 1;
 226                 else
 227                         tb->fastreuse = 0;
 228                 if (sk->sk_reuseport) {
 229                         tb->fastreuseport = 1;
 230                         tb->fastuid = uid;
 231                 } else
 232                         tb->fastreuseport = 0;
 233         } else {
 234                 if (tb->fastreuse &&
 235                     (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 236                         tb->fastreuse = 0;
 237                 if (tb->fastreuseport &&
 238                     (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
 239                         tb->fastreuseport = 0;
 240         }
 241 success:
 242         if (!inet_csk(sk)->icsk_bind_hash)
 243                 inet_bind_hash(sk, tb, snum);
 244         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 245         ret = 0;
 246
 247 fail_unlock:
 248         spin_unlock(&head->lock);
 249 fail:
 250         local_bh_enable();
 251         return ret;
 252 }
 253 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 254
 255 /*
 256  * Wait for an incoming connection, avoid race conditions. This must be called
 257  * with the socket locked.
 258  */
 259 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 260 {
 261         struct inet_connection_sock *icsk = inet_csk(sk);
 262         DEFINE_WAIT(wait);
 263         int err;
 264
 265         /*
 266          * True wake-one mechanism for incoming connections: only
 267          * one process gets woken up, not the 'whole herd'.
 268          * Since we do not 'race & poll' for established sockets
 269          * anymore, the common case will execute the loop only once.
 270          *
 271          * Subtle issue: "add_wait_queue_exclusive()" will be added
 272          * after any current non-exclusive waiters, and we know that
 273          * it will always _stay_ after any new non-exclusive waiters
 274          * because all non-exclusive waiters are added at the
 275          * beginning of the wait-queue. As such, it's ok to "drop"
 276          * our exclusiveness temporarily when we get woken up without
 277          * having to remove and re-insert us on the wait queue.
 278          */
 279         for (;;) {
 280                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 281                                           TASK_INTERRUPTIBLE);
 282                 release_sock(sk);
 283                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 284                         timeo = schedule_timeout(timeo);
 285                 lock_sock(sk);
 286                 err = 0;
 287                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 288                         break;
 289                 err = -EINVAL;
 290                 if (sk->sk_state != TCP_LISTEN)
 291                         break;
 292                 err = sock_intr_errno(timeo);
 293                 if (signal_pending(current))
 294                         break;
 295                 err = -EAGAIN;
 296                 if (!timeo)
 297                         break;
 298         }
 299         finish_wait(sk_sleep(sk), &wait);
 300         return err;
 301 }
 302
 303 /*
 304  * This will accept the next outstanding connection.
 305  */
 306 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 307 {
 308         struct inet_connection_sock *icsk = inet_csk(sk);
 309         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 310         struct sock *newsk;
 311         struct request_sock *req;
 312         int error;
 313
 314         lock_sock(sk);
 315
 316         /* We need to make sure that this socket is listening,
 317          * and that it has something pending.
 318          */
 319         error = -EINVAL;
 320         if (sk->sk_state != TCP_LISTEN)
 321                 goto out_err;
 322
 323         /* Find already established connection */
 324         if (reqsk_queue_empty(queue)) {
 325                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 326
 327                 /* If this is a non blocking socket don't sleep */
 328                 error = -EAGAIN;
 329                 if (!timeo)
 330                         goto out_err;
 331
 332                 error = inet_csk_wait_for_connect(sk, timeo);
 333                 if (error)
 334                         goto out_err;
 335         }
 336         req = reqsk_queue_remove(queue);
 337         newsk = req->sk;
 338
 339         sk_acceptq_removed(sk);
 340         if (sk->sk_protocol == IPPROTO_TCP && queue->fastopenq != NULL) {
 341                 spin_lock_bh(&queue->fastopenq->lock);
 342                 if (tcp_rsk(req)->listener) {
 343                         /* We are still waiting for the final ACK from 3WHS
 344                          * so can't free req now. Instead, we set req->sk to
 345                          * NULL to signify that the child socket is taken
 346                          * so reqsk_fastopen_remove() will free the req
 347                          * when 3WHS finishes (or is aborted).
 348                          */
 349                         req->sk = NULL;
 350                         req = NULL;
 351                 }
 352                 spin_unlock_bh(&queue->fastopenq->lock);
 353         }
 354 out:
 355         release_sock(sk);
 356         if (req)
 357                 __reqsk_free(req);
 358         return newsk;
 359 out_err:
 360         newsk = NULL;
 361         req = NULL;
 362         *err = error;
 363         goto out;
 364 }
 365 EXPORT_SYMBOL(inet_csk_accept);
 366
 367 /*
 368  * Using different timers for retransmit, delayed acks and probes
 369  * We may wish use just one timer maintaining a list of expire jiffies
 370  * to optimize.
 371  */
 372 void inet_csk_init_xmit_timers(struct sock *sk,
 373                                void (*retransmit_handler)(unsigned long),
 374                                void (*delack_handler)(unsigned long),
 375                                void (*keepalive_handler)(unsigned long))
 376 {
 377         struct inet_connection_sock *icsk = inet_csk(sk);
 378
 379         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 380                         (unsigned long)sk);
 381         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 382                         (unsigned long)sk);
 383         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 384         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 385 }
 386 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 387
 388 void inet_csk_clear_xmit_timers(struct sock *sk)
 389 {
 390         struct inet_connection_sock *icsk = inet_csk(sk);
 391
 392         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 393
 394         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 395         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 396         sk_stop_timer(sk, &sk->sk_timer);
 397 }
 398 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 399
 400 void inet_csk_delete_keepalive_timer(struct sock *sk)
 401 {
 402         sk_stop_timer(sk, &sk->sk_timer);
 403 }
 404 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 405
 406 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 407 {
 408         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 409 }
 410 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 411
 412 struct dst_entry *inet_csk_route_req(struct sock *sk,
 413                                      struct flowi4 *fl4,
 414                                      const struct request_sock *req)
 415 {
 416         struct rtable *rt;
 417         const struct inet_request_sock *ireq = inet_rsk(req);
 418         struct ip_options_rcu *opt = inet_rsk(req)->opt;
 419         struct net *net = sock_net(sk);
 420         int flags = inet_sk_flowi_flags(sk);
 421
 422         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 423                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 424                            sk->sk_protocol,
 425                            flags,
 426                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
 427                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
 428         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 429         rt = ip_route_output_flow(net, fl4, sk);
 430         if (IS_ERR(rt))
 431                 goto no_route;
 432         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 433                 goto route_err;
 434         return &rt->dst;
 435
 436 route_err:
 437         ip_rt_put(rt);
 438 no_route:
 439         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 440         return NULL;
 441 }
 442 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 443
 444 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
 445                                             struct sock *newsk,
 446                                             const struct request_sock *req)
 447 {
 448         const struct inet_request_sock *ireq = inet_rsk(req);
 449         struct inet_sock *newinet = inet_sk(newsk);
 450         struct ip_options_rcu *opt;
 451         struct net *net = sock_net(sk);
 452         struct flowi4 *fl4;
 453         struct rtable *rt;
 454
 455         fl4 = &newinet->cork.fl.u.ip4;
 456
 457         rcu_read_lock();
 458         opt = rcu_dereference(newinet->inet_opt);
 459         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 460                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 461                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 462                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
 463                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
 464         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 465         rt = ip_route_output_flow(net, fl4, sk);
 466         if (IS_ERR(rt))
 467                 goto no_route;
 468         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 469                 goto route_err;
 470         rcu_read_unlock();
 471         return &rt->dst;
 472
 473 route_err:
 474         ip_rt_put(rt);
 475 no_route:
 476         rcu_read_unlock();
 477         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 478         return NULL;
 479 }
 480 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 481
 482 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 483                                  const u32 rnd, const u32 synq_hsize)
 484 {
 485         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 486 }
 487
 488 #if IS_ENABLED(CONFIG_IPV6)
 489 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 490 #else
 491 #define AF_INET_FAMILY(fam) 1
 492 #endif
 493
 494 struct request_sock *inet_csk_search_req(const struct sock *sk,
 495                                          struct request_sock ***prevp,
 496                                          const __be16 rport, const __be32 raddr,
 497                                          const __be32 laddr)
 498 {
 499         const struct inet_connection_sock *icsk = inet_csk(sk);
 500         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 501         struct request_sock *req, **prev;
 502
 503         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 504                                                     lopt->nr_table_entries)];
 505              (req = *prev) != NULL;
 506              prev = &req->dl_next) {
 507                 const struct inet_request_sock *ireq = inet_rsk(req);
 508
 509                 if (ireq->rmt_port == rport &&
 510                     ireq->rmt_addr == raddr &&
 511                     ireq->loc_addr == laddr &&
 512                     AF_INET_FAMILY(req->rsk_ops->family)) {
 513                         WARN_ON(req->sk);
 514                         *prevp = prev;
 515                         break;
 516                 }
 517         }
 518
 519         return req;
 520 }
 521 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 522
 523 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 524                                    unsigned long timeout)
 525 {
 526         struct inet_connection_sock *icsk = inet_csk(sk);
 527         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 528         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 529                                      lopt->hash_rnd, lopt->nr_table_entries);
 530
 531         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 532         inet_csk_reqsk_queue_added(sk, timeout);
 533 }
 534 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 535
 536 /* Only thing we need from tcp.h */
 537 extern int sysctl_tcp_synack_retries;
 538
 539
 540 /* Decide when to expire the request and when to resend SYN-ACK */
 541 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 542                                   const int max_retries,
 543                                   const u8 rskq_defer_accept,
 544                                   int *expire, int *resend)
 545 {
 546         if (!rskq_defer_accept) {
 547                 *expire = req->num_timeout >= thresh;
 548                 *resend = 1;
 549                 return;
 550         }
 551         *expire = req->num_timeout >= thresh &&
 552                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 553         /*
 554          * Do not resend while waiting for data after ACK,
 555          * start to resend on end of deferring period to give
 556          * last chance for data or ACK to create established socket.
 557          */
 558         *resend = !inet_rsk(req)->acked ||
 559                   req->num_timeout >= rskq_defer_accept - 1;
 560 }
 561
 562 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
 563 {
 564         int err = req->rsk_ops->rtx_syn_ack(parent, req, NULL);
 565
 566         if (!err)
 567                 req->num_retrans++;
 568         return err;
 569 }
 570 EXPORT_SYMBOL(inet_rtx_syn_ack);
 571
 572 void inet_csk_reqsk_queue_prune(struct sock *parent,
 573                                 const unsigned long interval,
 574                                 const unsigned long timeout,
 575                                 const unsigned long max_rto)
 576 {
 577         struct inet_connection_sock *icsk = inet_csk(parent);
 578         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 579         struct listen_sock *lopt = queue->listen_opt;
 580         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 581         int thresh = max_retries;
 582         unsigned long now = jiffies;
 583         struct request_sock **reqp, *req;
 584         int i, budget;
 585
 586         if (lopt == NULL || lopt->qlen == 0)
 587                 return;
 588
 589         /* Normally all the openreqs are young and become mature
 590          * (i.e. converted to established socket) for first timeout.
 591          * If synack was not acknowledged for 1 second, it means
 592          * one of the following things: synack was lost, ack was lost,
 593          * rtt is high or nobody planned to ack (i.e. synflood).
 594          * When server is a bit loaded, queue is populated with old
 595          * open requests, reducing effective size of queue.
 596          * When server is well loaded, queue size reduces to zero
 597          * after several minutes of work. It is not synflood,
 598          * it is normal operation. The solution is pruning
 599          * too old entries overriding normal timeout, when
 600          * situation becomes dangerous.
 601          *
 602          * Essentially, we reserve half of room for young
 603          * embrions; and abort old ones without pity, if old
 604          * ones are about to clog our table.
 605          */
 606         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 607                 int young = (lopt->qlen_young<<1);
 608
 609                 while (thresh > 2) {
 610                         if (lopt->qlen < young)
 611                                 break;
 612                         thresh--;
 613                         young <<= 1;
 614                 }
 615         }
 616
 617         if (queue->rskq_defer_accept)
 618                 max_retries = queue->rskq_defer_accept;
 619
 620         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 621         i = lopt->clock_hand;
 622
 623         do {
 624                 reqp=&lopt->syn_table[i];
 625                 while ((req = *reqp) != NULL) {
 626                         if (time_after_eq(now, req->expires)) {
 627                                 int expire = 0, resend = 0;
 628
 629                                 syn_ack_recalc(req, thresh, max_retries,
 630                                                queue->rskq_defer_accept,
 631                                                &expire, &resend);
 632                                 req->rsk_ops->syn_ack_timeout(parent, req);
 633                                 if (!expire &&
 634                                     (!resend ||
 635                                      !inet_rtx_syn_ack(parent, req) ||
 636                                      inet_rsk(req)->acked)) {
 637                                         unsigned long timeo;
 638
 639                                         if (req->num_timeout++ == 0)
 640                                                 lopt->qlen_young--;
 641                                         timeo = min(timeout << req->num_timeout,
 642                                                     max_rto);
 643                                         req->expires = now + timeo;
 644                                         reqp = &req->dl_next;
 645                                         continue;
 646                                 }
 647
 648                                 /* Drop this request */
 649                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 650                                 reqsk_queue_removed(queue, req);
 651                                 reqsk_free(req);
 652                                 continue;
 653                         }
 654                         reqp = &req->dl_next;
 655                 }
 656
 657                 i = (i + 1) & (lopt->nr_table_entries - 1);
 658
 659         } while (--budget > 0);
 660
 661         lopt->clock_hand = i;
 662
 663         if (lopt->qlen)
 664                 inet_csk_reset_keepalive_timer(parent, interval);
 665 }
 666 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 667
 668 /**
 669  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 670  *      @sk: the socket to clone
 671  *      @req: request_sock
 672  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 673  *
 674  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 675  */
 676 struct sock *inet_csk_clone_lock(const struct sock *sk,
 677                                  const struct request_sock *req,
 678                                  const gfp_t priority)
 679 {
 680         struct sock *newsk = sk_clone_lock(sk, priority);
 681
 682         if (newsk != NULL) {
 683                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 684
 685                 newsk->sk_state = TCP_SYN_RECV;
 686                 newicsk->icsk_bind_hash = NULL;
 687
 688                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 689                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 690                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 691                 newsk->sk_write_space = sk_stream_write_space;
 692
 693                 newicsk->icsk_retransmits = 0;
 694                 newicsk->icsk_backoff     = 0;
 695                 newicsk->icsk_probes_out  = 0;
 696
 697                 /* Deinitialize accept_queue to trap illegal accesses. */
 698                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 699
 700                 security_inet_csk_clone(newsk, req);
 701         }
 702         return newsk;
 703 }
 704 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 705
 706 /*
 707  * At this point, there should be no process reference to this
 708  * socket, and thus no user references at all.  Therefore we
 709  * can assume the socket waitqueue is inactive and nobody will
 710  * try to jump onto it.
 711  */
 712 void inet_csk_destroy_sock(struct sock *sk)
 713 {
 714         WARN_ON(sk->sk_state != TCP_CLOSE);
 715         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 716
 717         /* It cannot be in hash table! */
 718         WARN_ON(!sk_unhashed(sk));
 719
 720         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 721         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 722
 723         sk->sk_prot->destroy(sk);
 724
 725         sk_stream_kill_queues(sk);
 726
 727         xfrm_sk_free_policy(sk);
 728
 729         sk_refcnt_debug_release(sk);
 730
 731         percpu_counter_dec(sk->sk_prot->orphan_count);
 732         sock_put(sk);
 733 }
 734 EXPORT_SYMBOL(inet_csk_destroy_sock);
 735
 736 /* This function allows to force a closure of a socket after the call to
 737  * tcp/dccp_create_openreq_child().
 738  */
 739 void inet_csk_prepare_forced_close(struct sock *sk)
 740 {
 741         /* sk_clone_lock locked the socket and set refcnt to 2 */
 742         bh_unlock_sock(sk);
 743         sock_put(sk);
 744
 745         /* The below has to be done to allow calling inet_csk_destroy_sock */
 746         sock_set_flag(sk, SOCK_DEAD);
 747         percpu_counter_inc(sk->sk_prot->orphan_count);
 748         inet_sk(sk)->inet_num = 0;
 749 }
 750 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 751
 752 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 753 {
 754         struct inet_sock *inet = inet_sk(sk);
 755         struct inet_connection_sock *icsk = inet_csk(sk);
 756         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 757
 758         if (rc != 0)
 759                 return rc;
 760
 761         sk->sk_max_ack_backlog = 0;
 762         sk->sk_ack_backlog = 0;
 763         inet_csk_delack_init(sk);
 764
 765         /* There is race window here: we announce ourselves listening,
 766          * but this transition is still not validated by get_port().
 767          * It is OK, because this socket enters to hash table only
 768          * after validation is complete.
 769          */
 770         sk->sk_state = TCP_LISTEN;
 771         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 772                 inet->inet_sport = htons(inet->inet_num);
 773
 774                 sk_dst_reset(sk);
 775                 sk->sk_prot->hash(sk);
 776
 777                 return 0;
 778         }
 779
 780         sk->sk_state = TCP_CLOSE;
 781         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 782         return -EADDRINUSE;
 783 }
 784 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 785
 786 /*
 787  *      This routine closes sockets which have been at least partially
 788  *      opened, but not yet accepted.
 789  */
 790 void inet_csk_listen_stop(struct sock *sk)
 791 {
 792         struct inet_connection_sock *icsk = inet_csk(sk);
 793         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 794         struct request_sock *acc_req;
 795         struct request_sock *req;
 796
 797         inet_csk_delete_keepalive_timer(sk);
 798
 799         /* make all the listen_opt local to us */
 800         acc_req = reqsk_queue_yank_acceptq(queue);
 801
 802         /* Following specs, it would be better either to send FIN
 803          * (and enter FIN-WAIT-1, it is normal close)
 804          * or to send active reset (abort).
 805          * Certainly, it is pretty dangerous while synflood, but it is
 806          * bad justification for our negligence 8)
 807          * To be honest, we are not able to make either
 808          * of the variants now.                 --ANK
 809          */
 810         reqsk_queue_destroy(queue);
 811
 812         while ((req = acc_req) != NULL) {
 813                 struct sock *child = req->sk;
 814
 815                 acc_req = req->dl_next;
 816
 817                 local_bh_disable();
 818                 bh_lock_sock(child);
 819                 WARN_ON(sock_owned_by_user(child));
 820                 sock_hold(child);
 821
 822                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 823
 824                 sock_orphan(child);
 825
 826                 percpu_counter_inc(sk->sk_prot->orphan_count);
 827
 828                 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->listener) {
 829                         BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 830                         BUG_ON(sk != tcp_rsk(req)->listener);
 831
 832                         /* Paranoid, to prevent race condition if
 833                          * an inbound pkt destined for child is
 834                          * blocked by sock lock in tcp_v4_rcv().
 835                          * Also to satisfy an assertion in
 836                          * tcp_v4_destroy_sock().
 837                          */
 838                         tcp_sk(child)->fastopen_rsk = NULL;
 839                         sock_put(sk);
 840                 }
 841                 inet_csk_destroy_sock(child);
 842
 843                 bh_unlock_sock(child);
 844                 local_bh_enable();
 845                 sock_put(child);
 846
 847                 sk_acceptq_removed(sk);
 848                 __reqsk_free(req);
 849         }
 850         if (queue->fastopenq != NULL) {
 851                 /* Free all the reqs queued in rskq_rst_head. */
 852                 spin_lock_bh(&queue->fastopenq->lock);
 853                 acc_req = queue->fastopenq->rskq_rst_head;
 854                 queue->fastopenq->rskq_rst_head = NULL;
 855                 spin_unlock_bh(&queue->fastopenq->lock);
 856                 while ((req = acc_req) != NULL) {
 857                         acc_req = req->dl_next;
 858                         __reqsk_free(req);
 859                 }
 860         }
 861         WARN_ON(sk->sk_ack_backlog);
 862 }
 863 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 864
 865 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 866 {
 867         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 868         const struct inet_sock *inet = inet_sk(sk);
 869
 870         sin->sin_family         = AF_INET;
 871         sin->sin_addr.s_addr    = inet->inet_daddr;
 872         sin->sin_port           = inet->inet_dport;
 873 }
 874 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 875
 876 #ifdef CONFIG_COMPAT
 877 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 878                                char __user *optval, int __user *optlen)
 879 {
 880         const struct inet_connection_sock *icsk = inet_csk(sk);
 881
 882         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 883                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 884                                                             optval, optlen);
 885         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 886                                              optval, optlen);
 887 }
 888 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 889
 890 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 891                                char __user *optval, unsigned int optlen)
 892 {
 893         const struct inet_connection_sock *icsk = inet_csk(sk);
 894
 895         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 896                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 897                                                             optval, optlen);
 898         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 899                                              optval, optlen);
 900 }
 901 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 902 #endif
 903
 904 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 905 {
 906         const struct inet_sock *inet = inet_sk(sk);
 907         const struct ip_options_rcu *inet_opt;
 908         __be32 daddr = inet->inet_daddr;
 909         struct flowi4 *fl4;
 910         struct rtable *rt;
 911
 912         rcu_read_lock();
 913         inet_opt = rcu_dereference(inet->inet_opt);
 914         if (inet_opt && inet_opt->opt.srr)
 915                 daddr = inet_opt->opt.faddr;
 916         fl4 = &fl->u.ip4;
 917         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 918                                    inet->inet_saddr, inet->inet_dport,
 919                                    inet->inet_sport, sk->sk_protocol,
 920                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 921         if (IS_ERR(rt))
 922                 rt = NULL;
 923         if (rt)
 924                 sk_setup_caps(sk, &rt->dst);
 925         rcu_read_unlock();
 926
 927         return &rt->dst;
 928 }
 929
 930 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 931 {
 932         struct dst_entry *dst = __sk_dst_check(sk, 0);
 933         struct inet_sock *inet = inet_sk(sk);
 934
 935         if (!dst) {
 936                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 937                 if (!dst)
 938                         goto out;
 939         }
 940         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 941
 942         dst = __sk_dst_check(sk, 0);
 943         if (!dst)
 944                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 945 out:
 946         return dst;
 947 }
 948 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);